Seattle, WA – Global supercomputer leader Cray has announced that the "Red Storm" supercomputer installed at Sandia National Laboratories in New Mexico is the first computer to surpass the 1 terabyte-per-second (1 TB/sec) performance mark on a widely used test that measures communications among processors in high-performance computing (HPC) systems and provides a key indication of the total communication capacity of the network.
Red Storm posted 1.8 TB/sec (1.8 trillion bytes per second) on the PTRANS interconnect bandwidth test that is part of the High Performance Computing Challenge (HPCC) test suite. By comparison, this figure represents 40 times more communications power per teraflop (trillion floating point operations per second) than the PTRANS result posted by an IBM Blue Gene system that has more than 10 times as many processors.
Red Storm was developed jointly by Sandia and Cray. The same architecture is used in the Cray XT3™ (AMD Opteron processors) massively parallel supercomputer, a system that is commercially available and has been installed in HPC centers around the world. During the low-workload December 2005 holiday period, Sandia was able to run tests on the entire 10,350-processor Red Storm system, which normally cannot be tested as one system because it is partitioned into classified and non-classified segments.
"We are delighted that Red Storm has broken the 1 TB/sec barrier on the HPCC PTRANS test," said Jeff Brooks, Cray's product manager for massively parallel processing. "While impressive teraflops results have become commonplace on Linpack-style tests, even on loosely connected computer clusters, Red Storm is the first system to achieve more than a terabyte per second on a communications-intensive test. This test is very important to HPC users because it is a good measure of overall system capability. It also indicates how well Cray supercomputers balance processing power and communications power. The results are right in line with the performance Cray users are seeing as they scale their demanding scientific applications to thousands of processors."